There is a fundamental gap in understanding how the retina responds under natural conditions. Over the last few decades, the discovery of nonlinear and long-range processes in the retina have made this gap particularly troublesome, since single-pathway linear models can not adequately explain the retina's behavior under natural stimuli. The long-term goal is to situate the retina within a theoretical framework that can explai and predict its responses to natural scenes. The overall objective in this application is to extend and test one theory, predictive coding, to provide a better explanation of the retina's behavior under natural scenes. The central hypothesis is that the retina implements predictive coding via inhibitory circuits that function to remove predictable - and hence redundant - features of the natural world. This hypothesis is substantiated by evidence that the retina appears to perform predictive coding under simple Gaussian stimuli as well as preliminary evidence from intracellular current injection experiments that show correlations of amacrine- ganglion cell pairs changing in a way consistent with predictive coding theory. The hypothesis will be tested with two specific aims that 1) extend the theory of predictive coding to natural scenes, and 2) make experimental tests of predictive coding by directly recording from and injecting current into inhibitory cells in the retina. Under the first aim, I will find nonlinear, biophysically feasible models of natural scenes that minimize prediction error; from the perspective of predictive coding, these models would form the ideal inhibitory circuit. I will then compare these models to the response properties of retinal ganglion cells and amacrine cells. The second aim makes use of intracellular current injection experiments to directly measure the prediction error between inhibitory cells and their downstream excitatory targets. This proposal is innovative, in the applicant's opinion, because it leverages both theoretical and experimental techniques to improve our framework for understanding the retina's behavior. Regardless of if the test of predictive coding is positive or negative, this study will generate new questions and resolve an ongoing controversy in the field. Additionally, this proposal will contribute to the understanding of retinal ganglion responses in natural environments that are essential to the development of next-generation retinal prosthetics.